Journal of Mathematical Modeling

Vol. 13, No. 1, 2025, pp. 139–152. Research Article 
JMM 

A novel fractional Bernoulli-Picard iteration method to solve

fractional differential equations
Fatemeh Zakipour, Abbas Saadatmandi∗
Department of Applied Mathematics, Faculty of Mathematical Sciences, University of Kashan,
Kashan 87317-53153, Iran
Email(s): f.zakipour70@gmail.com, saadatmandi@kashanu.ac.ir

Abstract. This paper introduces a new computational method for solving linear and non-linear frac-
tional differential equations (FDEs). Our method essentially consists of the combination of orthonormal
Bernoulli polynomials and the fractional form of the Picard iteration method. We name this method the
fractional Bernoulli-Picard iteration method (FBPIM). Unlike the spectral method, the proposed method
does not require solving a set of algebraic equations. We also discuss the convergence of the method.
Moreover, some numerical examples are included and compared with previously published results to
assess both the accuracy and suitability of the developed technique.

Keywords: Caputo derivative, Picard iteration method, orthonormal Bernoulli polynomials, fractional differential
equations, numerical solution.
AMS Subject Classification 2010: 26A33, 33C47.

1 Introduction
FDEs have gained significant attention in recent years due to their ability to describe numerous real-
world phenomena in various fields, including physics, engineering, biology, optimal control, and finance
(see e.g., [2,3,8,14] and references therein). Unlike ordinary differential equations, FDEs involve deriva-
tives of non-integer order, providing the ability to capture long-range dependencies and memory effects.
Solving FDEs presents a challenge as traditional analytical methods are often inadequate for this task.
So, over the past few decades, extensive research has been conducted on numerical approximation meth-
ods for solving FDEs. For example, we can refer to operational matrix method [12, 20, 21], collocation
method [11, 19], finite difference method [1] and other methods [10, 15].
This article focuses on combining two powerful methods for solving FDEs: orthonormal Bernoulli
∗ Corresponding author
Received: 01 September 2024/ Revised: 24 September 2024/ Accepted: 03 October 2024
DOI: 10.22124/jmm.2024.28329.2497

c 2025 University of Guilan http://jmm.guilan.ac.ir

140 F. Zakipour, A. Saadatmandi

polynomials (OBPs) approximation method and Picard iteration method. In recent years, various numer-
ical methods based on the Bernoulli polynomials have been developed. For example, in 2014, Tohidi
and his colleagues applied these polynomials to solve fractional integro-differential equations [27]. Ne-
mati [19] used the Bernoulli polynomial approximation method to obtain numerical solutions for FDEs.
In [13, 23], these polynomials are used for solving Fredholm-Volterra integro-differential equations of
fractional order. Most recently, Rahimkhani et al. proposed a technique based on the least square support
vector regression and Bernoulli wavelets for solving a system of FDEs [22]. Moreover, in [17], some
advantages of employing Bernoulli polynomials instead of Legendre polynomials to approximate a func-
tion are additionally mentioned. Although Bernoulli polynomials have very interesting properties, they
do not have orthogonality property. This property is particularly useful for many numerical methods.
Fortunately, an explicit representation of OBPs is given in [18].
On the other hand, the Picard method, also known as Picard iteration or successive approximation
method, constructs an infinite sequence of approximations to solve an equation. The core idea of the
Picard method is to reformulate differential equations into integral equations. Solving an integral equa-
tion iteratively yields a series of approximations that, under certain conditions (see e.g., [7]), converge
to the solution of the original differential equation. Note that it is not appropriate to use Picard’s method
directly as a numerical method, since computing integrals using Picard’s method can be very difficult
for nonlinear problems. Chebyshev-Picard iteration method was developed to solve this problem and is
often used to solve ordinary differential equations (see e.g., [5, 6]). Some new works in this area can also
be found in [25, 26].
To date, to our knowledge, only researchers have directly solved the FDE using Picard’s successive
approximation method (see e.g., [4, 16, 28]). In this paper, we generalize the Picard’s method for numer-
ically solving FDEs in a way that overcomes the drawbacks of using Picard’s successive approximation
method directly. This goal will be achieved by combining the Picard iteration method and the OBPs
approximation. This method is called FBPIM. Some advantages of FBPIM are that

(i) this method is easily implemented.

(ii) Despite several numerical techniques, in FBPIM, there is no need to solve a set of algebraic equa-
tions.

(iii) FBPIM does not require direct calculation of fractional derivatives and integrals.

The structure of this study is as follows. Section 2 reviews some preliminaries of fractional calculus and
recalling some properties related to Bernoulli polynomials. In Section 3, the fundamental formulation
of FBPIM is presented. The proof of convergence of FBPIM is given in Section 4, and some numerical
results are presented in Section 5. The conclusions are briefly summarized in Section 6.

2 Mathematical preliminaries
2.1 Preliminaries in fractional calculus
Both the Riemann-Liouville integral and Caputo derivative provide valuable tools in various fields of
science and engineering. Their usefulness lies in their ability to handle non-local or non-integer order
phenomena, providing a broader understanding and analysis of complex systems.
A novel FBPIM to solve FDEs 141

Definition 1. The Riemann-Liouville fractional integral with order α is defined as [8]

Z x
α 1
0 Ix z(x) = (x − λ )α−1 z(λ )dλ , x > 0, α > 0, (1)
Γ(α) 0
0
0 Ix z(x) = 1,

where Γ(·) is the Gamma function.

Definition 2. The Caputo fractional derivative of order α > 0 is defined as

C Dα z(x) 1 Rx n−α−1 (n)
0 x = Γ(n−α) 0 (x − λ ) z (λ )dλ , n − 1 < α < n, n ∈ N.

C Dα z(x)
0 x = z(n) (x), α = n.

The next section uses two important properties [8]

dαe−1
xi
α C α
0 Ix (0 Dx z(x)) = z(x) − ∑ z(i) (0) , x > 0, (2)
i=0 i!

α ζ Γ(ζ + 1) ζ +α
0 Ix x = x , α ≥ 0, ζ > −1, (3)
Γ(ζ + 1 + α)
where d·e is the ceiling function.

2.2 Preliminaries in OBPs

The well known Bernoulli polynomials Bn (x) are defined on the interval [0, 1] as
n  
n
Bn (x) = ∑ βν xn−ν , (4)
ν=0 ν

where the binomial coefficients are given by νn = ν!(n−ν)!

n!


and βν , ν = 0, . . . , n are Bernoulli numbers
[20]. As already mentioned, Bernoulli polynomials do not have the property of orthogonality. However,
their orthonormal form can be determined using the Gram-Schmidt orthonormalization method. An
explicit representation of OBPs on [0, 1] can be introduced as follows [18]
√ n   
ν n 2n − ν n−ν
Φn (x) = 2n + 1 ∑ (−1) x , n = 0, 1, 2, . . . . (5)
ν=0 ν n−ν

These polynomials create an orthonormal system over [0, 1]. Therefore, the function z(x) ∈ L2 ([0, 1]),
can be approximated as:
N
z(x) ≈ zN (x) = ∑ ei Φi (x) = EΦ(x), (6)
i=0

where the orthonormal Bernoulli coefficient vector E and orthonormal Bernoulli basis vector Φ(x) are
given by
E = [e0 , e1 , . . . , eN ], and Φ(x) = [Φ0 (x), Φ1 (x), . . . , ΦN (x)]T . (7)
142 F. Zakipour, A. Saadatmandi

Here, the unique coefficients {ei }Ni=0 can be computed as

Z 1
ei = z(x)Φi (x)dx, i = 0, 1, . . . , N.
0
Employing the Legendre-Gauss quadrature formula [21], ei can be approximated as follows
m
1
ei ≈ ∑ z(θ j )Φi (θ j )w j , i = 0, 1, . . . , N, (8)
2 j=0

where
1+tj 2
θj = , wj = 2 0 , j = 0, 1, . . . , m. (9)
2 (1 − t j )(Pm+1 (t j ))2
Here, Pm+1 (t),t ∈ [−1, 1] is the Legendre polynomial of order m + 1, and {t j }mj=0 are roots of Pm+1 (t).
Now, using the Taylor basis functions, the Bernoulli basis vector Φ(x) given in (7), can be expressed as
Φ(x) = BX(x), (10)
where X(x) = [1, x, x2 , . . . , xN ]T is the Taylor basis vector and B is the change-of-basis matrix. The entries
of matrix B can be computed using Eq. (5) as
  
0 0
 
(−1)0 0 ··· 0
 0  0 
√ √
    
1 1 1 2
3(−1)1 3(−1)0 ···
 
0
1 0 0 1
 
B= .
 
 .. .. .. .. 
. .
 . .
 
  
 √ 
N

N

√ N

N +1

√ N

2N 
2N + 1(−1)N 2N + 1(−1)N−1 ··· 2N + 1(−1)0
N 0 N −1 1 0 N

The matrix B is nonsingular. Therefore, Eq. (10) can be expressed as

X(x) = B−1 Φ(x). (11)
Applying 0 Ixα to Eq. (10), and using Eq. (3), we have
α
0 Ix Φ(x) = 0 Ixα (BX(x)) = B(0 Ixα X(x)) = BPxα X(x), (12)
where
1
···
 
Γ(1+α) 0 0 0 0
1

 0 Γ(2+α) 0 0 ··· 0 

2
0 0 0 ··· 0
 
P= Γ(3+α) .
.. .. .. .. ..
 
 
 . . . . . 
Γ(N+1)
0 0 0 0 ··· Γ(N+1+α)
By replacing Eq. (11) into Eq. (12), we get
α
0 Ix Φ(x) = BPxα B−1 Φ(x) = xα KΦ(x), (13)
where K = BPB−1 is an (N + 1) × (N + 1) lower triangular matrix.
Finally, note that to apply OBPs on any interval [a, b], we can use the following change of variable:
xb = (b − a)x + a, 0 ≤ x ≤ 1.
A novel FBPIM to solve FDEs 143

3 Description of the method

In this section, we will explain the basic idea of FBPIM for solving FDEs. Let us consider the following
single-term FDE
C α
0 Dx v(x) = g(x, v(x)), 0 ≤ x ≤ 1, 0 < α < 2, (14)

with standard initial conditions:

v(0) = va , v0 (0) = v0a . (15)
Here, the second initial condition holds only for α > 1. First, use the transformation z(x) = v(x)−va −v0a x
to convert the problem (14)-(15) into the following FDE with homogeneous initial conditions:
C α
0 Dx z(x) =h(x, z(x)), 0 ≤ x ≤ 1, 0 < α < 2, (16)
0
z(0) =0, z (0) = 0.

Next, applying the fractional integral operator 0 Ixα to both sides of the FDE given in Eq. (16) and using
Eq. (2), we get Z x
1
z(x) = (x − λ )α−1 h(λ , z(λ ))dλ . (17)
Γ(α) 0
Therefore, we can generate a sequence of Picard iterations as follows:
Z x
1
zi (x) = (x − λ )α−1 h(λ , zi−1 (λ ))dλ i = 0, 1, . . . , (18)
Γ(α) 0

where, zi−1 (x) stands for the approximate solution at the (i − 1)th step and z0 (x) is an essential initial
i ∞
guess. Convergence of sequence z (x) i=0 to z(x) is discussed by [9, 28]. Now, we numerically im-
plement Picard iteration based on the OBPs for Eq. (18). For this purpose, first we use Eq. (6) to
approximate the function h(λ , zi−1 (λ )) by OBPs as:
N
h(λ , zi−1 (λ )) ≈ ∑ hri−1 Φr (λ ) = H i−1 Φ(λ ), (19)
r=0

where H i−1 = [hi−1 i−1 i−1 i−1 N

0 , h1 , . . . , hN ] is a coefficient vector. The coefficients {hr }r=0 can be calculated
from Eq. (8) as:
m
1
hi−1
r ≈ ∑ h(θ j , zi−1 (θ j ))Φr (θ j )w j , r = 0, . . . , N. (20)
2 j=0

Rewriting Eq. (20) as a matrix form, we obtain

H i−1 = h(θ0 , zi−1 (θ0 )), h(θ1 , zi−1 (θ1 )), . . . , h(θm , zi−1 (θm ))
 
 1  
2 w0 · · · 0 Φ0 (θ0 ) · · · ΦN (θ0 )
×  ... .. ..   .. .. .. i−1
 = F WQ, (21)
 
. .  . . .
1
0 ··· 2 wm Φ0 (θm ) · · · ΦN (θm )
144 F. Zakipour, A. Saadatmandi

where

F i−1 = h(θ0 , zi−1 (θ0 )), h(θ1 , zi−1 (θ1 )), . . . , h(θm , zi−1 (θm )) ,
 
   1 
Φ0 (θ0 ) · · · ΦN (θ0 ) 2 w0 . . . 0
.
.. . .. .
.. . . ..
Q= , W =  .. .. . (22)
   
.
1
Φ0 (θm ) · · · ΦN (θm ) 0 ··· 2 wm

Note that matrices Q and W are independent of i, and the function h, so they only need to be calculated
once at the beginning of the algorithm. The subsequent approximation is obtained by substituting Eq.
(19) into Eq. (18):
!
Z x N
1
zi (x) ≈ (x − λ )α−1 ∑ hi−1 r Φr (λ ) dλ
Γ(α) 0 r=0
N  Z x  N
i−1 1
= ∑ hr (x − λ ) Φr (λ )dλ = ∑ hi−1
α−1 α
r (0 Ix Φr (x)). (23)
r=0 Γ(α) 0 r=0

Now, we approximate zi (x) in the following form:

N
zi (x) ≈ xα ∑ zik Φk (x) = xα Z i Φ(x), (24)
k=0

where Z i = [zi0 , zi1 , . . . , ziN ] is an unknown vector. Using Eqs. (13), (19) and (24), we can rewrite Eq. (23)
in the following vector-matrix form:

xα Z i Φ(x) = xα H i−1 KΦ(x). (25)

Therefore,
Z i = H i−1 K, (26)
where H i−1 is calculated using Eq. (21). Once the vector Z i is found, the approximate solution zi (x)
can be obtained from Eq. (24). The iterative process continues until the desired level of accuracy in
∞
the approximate solutions is achieved. For example, the process of creating sequence zi (x) i=0 can
continue until:

max{|zi (θ j ) − zi−1 (θ j )|, j = 0, . . . , m} < ε, (27)

where ε is a given tolerance. As you can see, there is no need to solve any set of algebraic equations to
find zi (x). This is one of the most important advantages of FBPIM.

4 Convergence analysis
In this section, we discuss the convergence of FBPIM. It is important to note that, due to the use of
a truncated series of OBPs at each stage and the accumulation of rounding errors, the actual domain
of convergence of the FBPIM may differ from the convergence domain of the classical Picard iteration
method. The norm used in this section is the maximum norm, i.e., k · k = k · k∞ . Also, let J = [0, 1],
A novel FBPIM to solve FDEs 145

R+ = [0, ∞) and C(J) be the Banach space of all real-valued continuous functions on J.
In sequel, we consider fractional order integral equation (17) under the following assumptions:
(i) h on J × R+ is continuous and bounded with M = sup |h(x, z)|.
(x,z)∈J×R+
(ii) h satisfies the Lipschitz condition as

|h(t, z1 ) − h(t, z2 )| ≤ L|z1 − z2 |,

where L is a real positive constant.

Lemma 1. ( [9, Theorem 2.1]). Assume (i) and (ii) hold. If LM < 1, then (17) has a unique positive
solution z ∈ C(J).
Lemma 2. ( [9]). Let us assume that the conditions of Lemma 1 hold. In this case, all functions zi (x)
i ∞
obtained from Eq. (18) are continuous and the sequence z (x) i=0 uniformly converges to z(x) ∈ C(J).
Here z(x) is the exact solution of (17).
Lemma 3. Let Φ(x) be the orhonormal Bernoulli basis vector on the interval [0, 1]. Then kΦ(x)k ≤ kBk,
where B is given in (10).
Proof. Using Eq. (10), we have kΦ(x)k ≤ kBk kX(x)k = kBk.
| {z }
1

Theorem 1. Let us assume that the conditions of Lemma 2 hold. If ∆ = LkWkkQkKk||B|| < 1, then the
iterative scheme (18) for the FDE (16) using FBPIM converges.
Proof. All functions zi (x) are continuous, and zi (x) can be expressed as the sum of consecutive differ-
ences
n
zi (x) = z0 (x) + ∑ (zk (x) − zk−1 (x)).
k=1
∞
(zk (x) − zk−1 (x))
is convergent then the sequence zi (x)

Therefore, if the infinite series ∑ i=0
will con-
verge to z(x). Now, using Eqs. (26) and (21), we obtain

Z k − Z k−1 = H k−1 K − H k−2 K = (H k−1 − H k−2 )K

= (F k−1 − F k−2 )WQK. (28)

Then, applying the triangle inequality and the Lipschitz condition, we get

kZ k − Z k−1 k ≤ kF k−1 − F k−2 k||W||||Q||||K||

= max |h(θ j , zk−1 (θ j )) − h(θ j , zk−2 (θ j ))|||W||||Q||K||
0≤ j≤m

≤ L max |zk−1 (θ j ) − zk−2 (θ j )|||W||||Q||K||

0≤ j≤m

≤ L||zk−1 (x) − zk−2 (x)||||W||||Q||||K||. (29)

Using Eq. (24) and Lemma 3, we have

kzk (x) − zk−1 (x)k ≤ kxα kkZ k − Z k−1 k||Φ(x)|| ≤ kZ k − Z k−1 k||B||. (30)
146 F. Zakipour, A. Saadatmandi

From (29) and (30), we obtain

||zk (x) − zk−1 (x)|| ≤ ||zk−1 (x) − zk−2 (x)||∆

≤ ||zk−2 (x) − zk−3 (x)||∆2
..
.
≤ ||z1 (x) − z0 (x)||∆k−1 . (31)

Since ∆ < 1, then the uniform convergence of ∑∞ k k−1 (x)) is proved and as a result lim zi (x) =
k=1 (z (x) − z
i→∞
z(x).

5 Illustrative examples
In this section, to verify the performance and accuracy of the FBPIM for solving FDEs, we present
several examples. Our computations were performed by Maple programming on a personal computer
equipped with 3.40 GHz, Core i7, and 4 GB of memory.

Example 1. Consider the following linear FDE [24]:

C α
0 Dx z(x) + z(x) = 0, 0 < α ≤ 2, (32)

with initial conditions

z(0) = 1, if 0 < α < 1, (33)
or
z(0) = 1, z0 (0) = 0, if 1 < α < 2. (34)

The analytical solution is z(x) = Eα (−xα ), in which Eα (x) is the Mittag-Leffler function defined by
∞
xk
Eα (x) = ∑ Γ(kα + 1) , α > 0.
k=0

Eq. (32) is referred to as the simple fractional relaxation equation and the simple fractional oscillation
equation for 0 < α < 1 and 1 < α < 2 respectively [10]. The absolute errors obtained by utilizing the
FBPIM, described in Section 3, with N = m = 10, ε = 10−13 and for different values of α are illustrated
in Table 1. Also, in Table 1, we have compared the absolute error of the FBPIM and the Legendre
operational matrix method given in [24]. Indeed, the plotted numerical results for z(x) are depicted in
Figure 1, which displays the outcomes for N = m = 20, and different values of α, namely 0.5, 0.75, and
0.95. It is worth noting that when α = 1, the analytical solution can be expressed as z(x) = exp(−x).
Figure 1 illustrates that as the parameter α approaches 1, the numerical solution gradually converges
towards the analytical solution z(x) = exp(−x). On the other hand, when α = 2, the analytical solution
becomes z(x) = cos(x). The numerical results for z(x) with N = m = 20 and α = 1.5, 1.75, 1.95 are
displayed in Figure 2. Once again, it can be observed from Figure 2 that as α approaches 2, the numerical
solution converges to the analytical solution z(x) = cos(x). These Figures demonstrate the corresponding
solutions of the FDE approach to the solutions of integer order differential equation.
A novel FBPIM to solve FDEs 147

Table 1: Absolute error for different values of α, N = m = 10 and ε = 10−13 for Example 1.

α = 0.6 α = 1.2 α = 1.6

x Method [24] FBPIM Method [24] FBPIM Method [24] FBPIM
0.1 6.7 ×10−3 9.4×10−5 1.3 ×10−3 3.5×10−7 3.0×10−4 1.9×10−8
0.3 2.0 ×10−5 2.1×10−6 2.8 ×10−3 5.1×10−7 1.3×10−4 1.1×10−7
0.5 5.2 ×10−3 4.8×10−6 4.5 ×10−3 1.0×10−6 3.1×10−4 2.1×10−7
0.7 4.4 ×10−3 3.8×10−5 3.6 ×10−3 1.0×10−6 3.0×10−4 2.1×10−7
0.9 4.6 ×10−3 2.8×10−5 1.8 ×10−3 7.2×10−7 6.2×10−7 2.0×10−7

exp(-x)
0.9 = 0.5
= 0.75
= 0.95

0.8

0.7
z(x)

0.6

0.5

0.4

0.3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Figure 1: Numerical solutions for α = 0.5, 0.75, 0.95, and with N = m = 20 and ε = 10−25 , for Example 1.

Example 2. Consider the following non-linear FDE [8, 24]

C α 40320 8−α Γ(5 + α/2) 4−α/2 9

0 Dx z(x) = x −3 x + Γ(α + 1) (35)
Γ(9 − α) Γ(5 − α/2) 4
 3
3 α/2
+ x − x4 − z(x)3/2 ,
2
with initial conditions
z(0) = 0, if 0 < α < 1, (36)
or
z(0) = 0, z0 (0) = 0, if 1 < α < 2. (37)
The analytical solution is given by:
9
z(x) = x8 − 3x4+α/2 + xα . (38)
4
148 F. Zakipour, A. Saadatmandi

cos(x)
= 1.5
= 1.75
0.9 = 1.95

0.8

0.7
z(x)

0.6

0.5

0.4

0.3
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x

Figure 2: Numerical solutions for α = 1.5, 1.75, 1.95, N = m = 20 and ε = 10−25 for Example 1.

Table 2: Absolute error for different values of α, N = m = 10 and ε = 10−10 for Example 2.

α = 0.4 α = 0.8 α = 1.4

x Method [24] FBPIM Method [24] FBPIM Method [24] FBPIM
0.1 6.3 ×10−2 9.4×10−8 2.9 ×10−3 7.0×10−9 2.0×10−4 5.8×10−8
0.3 6.0 ×10−2 2.5×10−8 2.1×10−3 9.1×10−8 1.6×10−3 5.2×10−8
0.5 2.4 ×10−2 5.3×10−8 2.3×10−3 1.9×10−8 7.6×10−3 4.7×10−8
0.7 1.2 ×10−1 4.6×10−8 2.5×10−3 6.9×10−8 4.9×10−3 3.1×10−8
0.9 3.0 ×10−1 4.3×10−8 2.1 ×10−3 2.9×10−8 3.3×10−2 7.3×10−9

The Table 2, presents the absolute errors acquired through the utilization of the FBPIM, as explained
in Section 3, with N = m = 10, ε = 10−10 , and various values of α. Additionally, the absolute error
of the FBPIM is compared to the Legendre operational matrix method provided in [24] within the same
Table 2. This table illustrates that our methodology yields a numerical solution that closely approximates
the exact solution in comparison to the technique proposed in [24]. Also, Figures 3 and 4 illustrate the
graphical representation of the absolute error functions for α = 0.7, 1.5, and different values of N = m.
Example 3. Consider the following fractional Riccati differential equation [29]:
C α
0 Dx z(x) = −z2 (x) + 2z(x) + 1, 0 < α ≤ 1, (39)
with initial condition z(0) = 0. The analytical solution for α = 1 is given by:
√ √ √ 
z(x) = 1 + 2 tanh 2x + ln( 2 − 1) .

The graphical depiction of the absolute error functions for α = 1 and varying values of N = m are
illustrated in Figure 5. This figure demonstrates a rapid decrease in absolute errors as both N and m
A novel FBPIM to solve FDEs 149

Figure 3: Graph of absolute error functions with Figure 4: Graph of absolute error functions with
α = 0.7, ε = 10−10 and for various values of N = m, α = 1.5, ε = 10−10 and for various values of N = m,
for Example 2. for Example 2.

Figure 5: Graph of absolute error functions with Figure 6: Numerical solutions with α =
α = 1, ε = 10−15 and for various values of N = m, 0.5, 0.75, 0.9, 1, N = m = 10 and ε = 10−10 , for Ex-
for Example 3. ample 3.

are increased. Also, the results for different values of α are shown in Figure 6. Figure 6 shows the
convergence of the numerical solution to the analytical solution as α approaches 1.
Furthermore, the maximum absolute error (denoted by L∞ −error) and CPU time (in seconds) of
FBPIM for all three examples and various values of N = m are presented in Table 3. By increasing the
value of N, it can be clearly observed from Table 3 that the L∞ −errors decrease.
150 F. Zakipour, A. Saadatmandi

Table 3: Maximum absolute error and CPU time for different values of N = m and ε = 10−15 .

Example 1 (α = 0.8) Example 2 (α = 0.8) Example 3 (α = 1)

N L∞ −error CPU time L∞ −error CPU time L∞ −error CPU time
7 2.0×10−4 1.467 7.3 ×10−5 0.593 1.8×10−5 0.812
10 6.2×10−5 1.778 2.2 ×10−7 0.718 1.2×10−7 1.046
15 1.8×10−5 3.120 5.9 ×10−9 0.936 9.3×10−11 1.841
20 7.2×10−6 7.707 4.5 ×10−10 1.607 2.9×10−14 3.807

6 Conclusion
In this paper, we have developed the FBPIM for solving FDEs. Our method distinguishes itself from
other methods such as Galerkin and collocation methods by avoiding the need to solve a system of alge-
braic equations in order to determine the solution. The FBPIM has been effectively utilized for solving
both linear and non-linear FDEs. In our computational investigations, we have presented the outcomes
derived by utilizing the FBPIM on different FDEs. In general, the FBPIM presents a significant method
for computationally resolving FDEs and holds promise for broader implementation in the discipline.

References
[1] M. Abbaszadeh, A. Bagheri Salec, A. Al-Khafaji, S. Kamel, The effect of fractional-order deriva-
tive for pattern formation of Brusselator reaction-diffusion model occurring in chemical reactions,
Iranian J. Math. Chem. 14 (2023) 243–269.
[2] W.M. Abd-Elhameed, Y.H. Youssri, A.G. Atta, Tau algorithm for fractional delay differential equa-
tions utilizing seventh-kind Chebyshev polynomials, J. Math. Model. 12 (2024) 277–299.
[3] R. Akbari, L. Navaei, Fractional dynamics of infectious disease transmission with optimal control,
Math. Interdisc. Res. 9 (2024) 199–213.
[4] R. Almeida, A.B. Malinowska, M.T.T. Monteiro, Fractional differential equations with a Caputo
derivative with respect to a Kernel function and their applications, Math. Methods Appl. Sci. 41
(2018) 336–352.
[5] X. Bai, J.L. Junkins, Modified Chebyshev-Picard iteration methods for solution of boundary value
problems, J. Astronaut. Sci. 58 (2011) 615–642.
[6] X. Bai, J.L. Junkins, Modified Chebyshev-Picard iteration methods for solution of initial value
problems, J. Astronaut. Sci. 59 (2012) 327–351.
[7] W.J. Coles, T.L. Sherman, Convergence of successive approximations for nonlinear two-point
boundary value problems, SIAM J. Appl. Math. 15 (1967) 426–433.
[8] K. Diethelm, The Analysis of Fractional Differential Equations, Springer-Verlag, Berlin, 2010.
[9] A.M.A. El-Sayed, H.H.G. Hashem, E.A.A. Ziada, Picard and Adomian decomposition methods for
a quadratic integral equation of fractional order, Comput. Appl. Math. 33 (2014) 95–109.
A novel FBPIM to solve FDEs 151

[10] S. Esmaeili, M. Shamsi, M. Dehghan, Numerical solution of fractional differential equations via a
Volterra integral equation approach, Open Phys. 11 (2013) 1470–1481.

[11] S. Esmaeili, M. Shamsi, Y. Luchko, Numerical solution of fractional differential equations with a
collocation method based on Müntz polynomials, Comput. Appl. Math, 62 (2011) 918–929.

[12] E. Keshavarz, Y. Ordokhani, M. Razzaghi, Bernoulli wavelet operational matrix of fractional order
integration and its applications in solving the fractional order differential equations, Appl. Math.
Modelling, 38 (2014) 6038–6051.

[13] E. Keshavarz, Y. Ordokhani, M. Razzaghi, Numerical solution of nonlinear mixed Fredholm-

Volterra integro-differential equations of fractional order by Bernoulli wavelets, Comput. Methods
Differ. Equ. 7 (2019) 163–176.

[14] A.A. Kilbas, H.M. Srivastava, J.J. Trujillo, Theory and Applications of Fractional Differential
Equations, Elsevier, San Diego, 2006.

[15] P. Kumar, O.P. Agrawal, An approximate method for numerical solution of fractional differential
equations, Signal Process. 86 (2006) 2602–2610.

[16] R. Lyons, A.S. Vatsala, R.A. Chiquet, Picard’s iterative method for Caputo fractional differential
equations with numerical results, Mathematics, 5 (2017) 65.

[17] S. Mashayekhi, Y. Ordokhani, M. Razzaghi, Hybrid functions approach for nonlinear constrained
optimal control problems, Commun. Nonlinear Sci. Numer. Simul. 17 (2012) 1831–1843.

[18] F. Mirzaee, N. Samadyar, Explicit representation of orthonormal Bernoulli polynomials and its
application for solving Volterra-Fredholm-Hammerstein integral equations, SeMA J. 77 (2020)
81–96.

[19] S. Nemati, P.M. Lima, D.F. Torres, Numerical solution of variable-order fractional differential
equations using Bernoulli polynomials, Fractal fract. 5 (2021) 219.

[20] M. Pourbabaee, A. Saadatmandi, New operational matrix of Riemann-Liouville fractional deriva-

tive of orthonormal Bernoulli polynomials for the numerical solution of some distributed-order
time-fractional partial differential equations, J. Appl. Anal. Comput. 13 (2023) 3352–3373.

[21] M. Pourbabaee, A. Saadatmandi, A new operational matrix based on Müntz-Legendre polynomials

for solving distributed order fractional differential equations, Math. Comput. Simulat. 194 (2022)
210–235.

[22] P. Rahimkhani, Y. Ordokhani, S. Sabermahani, Bernoulli wavelet least square support vector re-
gression: Robust numerical method for systems of fractional differential equations, Math. Methods
Appl. Sci. 46 (2023) 17641–17659.

[23] A. Saadatmandi, S. Akhlaghi, Using hybrid of block-pulse functions and Bernoulli polynomials to
solve fractional Fredholm-Volterra integro-differential equations, Sains Malays. 49 (2020) 953–
962.
152 F. Zakipour, A. Saadatmandi

[24] A. Saadatmandi, M. Dehghan, A new operational matrix for solving fractional-order differential
equations, Comput. Math. Appl. 59 (2010) 1326–1336.

[25] M. Tafakkori-Bafghi, G.B. Loghmani, M. Heydari, Numerical solution of two-point nonlinear

boundary value problems via Legendre-Picard iteration method, Math. Comput. Simul. 199 (2022)
133–159.

[26] M. Tafakkori-Bafghi, G.B. Loghmani, M. Heydari, X. Bai, Jacobi-Picard iteration method for the
numerical solution of nonlinear initial value problems, Math. Methods Appl. Sci. 43 (2020) 1084–
1111.

[27] E. Tohidi, M.M. Ezadkhah, S. Shateyi, Numerical solution of nonlinear fractional Volterra integro-
differential equations via Bernoulli polynomials, Abstr. Appl. Anal. 2014 (2014) 162896.

[28] X. Yang, Y. Liu, Picard iterative processes for initial value problems of singular fractional differ-
ential equations, Adv. Differ. Equ. 2014 (2014) 102.

[29] Ş. Yüzbaşı, Numerical solutions of fractional Riccati type differential equations by means of the
Bernstein polynomials, Appl. Math. Comput. 219 (2013) 6328–6343.